A method, device and computer equipment for generating face annotation data
Through the similarity calculation and weighting processing of multiple face recognition algorithms, combined with clustering and exception filtering, the problem of lack of generalization ability in the generation method of face labeling data in the existing technology is solved, and more accurate face quality evaluation is achieved and better model training effects.
Patent Information
- Application Number
- CN202411109423.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-13
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2044-08-13
AI Technical Summary
The existing face labeling data generation methods rely too much on the current face recognition algorithm, resulting in a lack of generalization ability and inability to effectively evaluate face quality.
By obtaining the face image to be marked and its corresponding standard images, using preset multiple face recognition algorithms to calculate the similarity, and weighted calculations based on the correlation between different algorithms to generate comprehensive annotation results. Effective labeled data are obtained through clustering and exception filtering.
Reliance on a single face recognition algorithm is reduced, and the generalization ability of face labeling data is improved. The generated labeling data can more accurately evaluate face quality, and enhance the training effect of the face quality evaluation model.
Smart Images

Figure CN118644749B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of face recognition technology, and in particular to a method, device and computer equipment for generating face annotation data. Background Art
[0002] With the development of electronic technology, face recognition technology has been gradually applied to many fields. With the widespread application of face recognition technology, people's requirements for the accuracy of face recognition are getting higher and higher. Among them, the accuracy of face quality assessment plays a key role in the accuracy of face recognition.
[0003] In traditional technology, a face quality assessment model is trained mainly through a large amount of face annotation data to obtain a fully trained face quality assessment model. Through the trained face quality assessment model, the face quality of the image for face recognition is assessed. Among them, the acquisition of a large amount of effective face annotation data is the key to obtaining an accurate face quality assessment model.
[0004] The existing method of generating face annotation data usually uses the target face image similarity distribution distance to score the face quality of the target face image and obtain the target face annotation data. However, the existing method of generating face annotation data has a great impact on the accuracy of the quality score of the face annotation data due to the number and quality distribution of images in the category. It is overly dependent on the current face recognition algorithm and does not have generalization ability.
[0005] However, there is still no effective solution to the problem that the existing method of generating face annotation data relies too much on the current face recognition algorithm and lacks generalization ability. Summary of the invention
[0006] Based on this, it is necessary to provide a method, device and computer equipment for generating face annotation data to address the above technical problems.
[0007] In a first aspect, the present application provides a method for generating face annotation data. The method comprises the following steps:
[0008] Acquire a face image to be annotated and a standard image corresponding to the face image to be annotated;
[0009] Using a plurality of preset face recognition algorithms, similarity calculation is performed on each of the face images to be labeled and the face features in the corresponding standard image, so as to obtain a first labeling result of each of the face images to be labeled for each face recognition algorithm;
[0010] Based on the correlation between each of the face recognition algorithms and other face recognition algorithms in the preset multiple face recognition algorithms, weighted calculation is performed on each of the face images to be labeled for the first labeling results of each of the face recognition algorithms to obtain a second labeling result for each of the face images to be labeled.
[0011] In one embodiment, the method for generating face annotation data further includes the following steps:
[0012] Abnormal data in the second annotation results of each of the face images to be annotated are filtered to obtain valid face annotation data.
[0013] In one embodiment, the method of using a plurality of preset face recognition algorithms to perform similarity calculation on each of the face images to be labeled and the face features in the corresponding standard image to obtain a first labeling result of each of the face images to be labeled for each face recognition algorithm includes the following steps:
[0014] Using the preset multiple face recognition algorithms, similarity calculation is performed on each of the face images to be labeled and the face features in the corresponding standard image to obtain similarity results of each of the face images to be labeled under each of the face recognition algorithms;
[0015] The similarity results of each of the facial images to be labeled under each of the facial recognition algorithms are expressed in the same measurement method to obtain the first labeling results of each of the facial images to be labeled for each of the facial recognition algorithms.
[0016] In one embodiment, based on the correlation between each face recognition algorithm and other face recognition algorithms in the preset multiple face recognition algorithms, weighted calculation is performed on each face image to be labeled for the first labeling result of each face recognition algorithm to obtain the second labeling result of each face image to be labeled, including the following steps:
[0017] Based on the correlation between each face recognition algorithm and other face recognition algorithms in the preset multiple face recognition algorithms, a weighting coefficient of each face image to be labeled under each face recognition algorithm is obtained;
[0018] Based on the weighted coefficients of each of the facial images to be labeled under each of the facial recognition algorithms, weighted calculations are performed on each of the facial images to be labeled for the first labeling results of each of the facial recognition algorithms to obtain second labeling results for each of the facial images to be labeled.
[0019] In one embodiment, the step of obtaining the weighting coefficient of each face image to be labeled under each face recognition algorithm based on the correlation between each face recognition algorithm and other face recognition algorithms in the preset multiple face recognition algorithms comprises the following steps:
[0020] Based on the correlation between each face recognition algorithm and other face recognition algorithms in the preset multiple face recognition algorithms, obtaining the average value of the correlation between each face recognition algorithm and other face recognition algorithms in the preset multiple face recognition algorithms;
[0021] The normalized result of the mean value of the correlation between the current face recognition algorithm and other face recognition algorithms in the preset multiple face recognition algorithms is used as the weighting coefficient of each face image to be labeled under the current face recognition algorithm.
[0022] In one embodiment, after obtaining the second annotation result of each of the face images to be annotated, the following steps are also included:
[0023] Based on the preset absolute quality scores of the facial images to be labeled, and in accordance with the preset step length, segmentally mapping the second labeling results of the facial images to be labeled, to obtain third labeling results of the facial images to be labeled;
[0024] Abnormal data in the third annotation result of each of the face images to be annotated is filtered to obtain the valid face annotation data.
[0025] In one embodiment, filtering abnormal data in the second annotation results of each of the face images to be annotated to obtain valid face annotation data includes the following steps:
[0026] Clustering the face images to be labeled according to face features to obtain clustering results;
[0027] Based on the number of facial images to be labeled in each cluster of the clustering result and the second labeling result of each facial image to be labeled, simplifying the facial images to be labeled in each cluster to obtain simplified facial images to be labeled;
[0028] The streamlined face images to be annotated are sorted according to the second annotation results, and neighborhood anomaly filtering is performed on the sorted second annotation results of the face images to be annotated to obtain the valid face annotation data.
[0029] In one embodiment, based on the number of facial images to be labeled in each cluster of the clustering result and the second labeling result of each facial image to be labeled, the facial images to be labeled in each cluster are simplified to obtain the simplified facial images to be labeled, including the following steps:
[0030] When the number of the facial images to be labeled in the cluster is greater than or equal to 2, calculating the difference between the second labeling results of each of the facial images to be labeled in the cluster;
[0031] Based on the difference between the second annotation results of each of the facial images to be annotated in the cluster, the facial images to be annotated in each of the clusters are simplified to obtain the simplified facial images to be annotated.
[0032] In one embodiment, the process of sorting the streamlined facial images to be annotated according to the second annotation results, and performing neighborhood anomaly filtering on the sorted second annotation results of the facial images to be annotated to obtain the valid facial annotation data includes the following steps:
[0033] Sorting the streamlined facial images to be annotated according to the second annotation result, and calculating the similarity of adjacent images and interval images for facial features in the sorted facial images to be annotated, to obtain similarity results of adjacent images and similarity results of interval images;
[0034] Based on the similarity results of the adjacent images and the similarity results of the interval images, the simplified face image is subjected to feature anomaly filtering to obtain a face image to be identified after feature anomaly filtering;
[0035] Based on the difference in attribute scores between adjacent facial images to be identified after feature anomaly filtering and a preset attribute score difference threshold, attribute anomaly filtering is performed on the facial images to be identified after feature anomaly filtering to obtain the valid facial annotation data.
[0036] In a second aspect, the present application also provides a device for generating face annotation data. The device comprises:
[0037] An acquisition module, used to acquire a face image to be annotated and a standard image corresponding to the face image to be annotated;
[0038] A first calculation module is used to use a plurality of preset face recognition algorithms to perform similarity calculations on the face features in each of the face images to be labeled and the corresponding standard images, so as to obtain a first labeling result of each of the face images to be labeled for each face recognition algorithm;
[0039] And a second calculation module is used to perform weighted calculation on each of the face images to be labeled for the first labeling results of each of the face recognition algorithms based on the correlation between each of the face recognition algorithms and other face recognition algorithms in the preset multiple face recognition algorithms, so as to obtain a second labeling result for each of the face images to be labeled.
[0040] In a third aspect, the present application further provides a computer device, wherein the computer device comprises a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the method for generating face annotation data described in the first aspect are implemented.
[0041] The above-mentioned method, device and computer equipment for generating face annotation data calculate the similarity between each face image to be annotated and the face features in the standard image corresponding to the face image to be annotated by using multiple preset face recognition algorithms, and obtain the first annotation result of each face image to be annotated for each face recognition algorithm, and then perform weighted calculation on the first annotation result of each face image to be annotated for each face recognition algorithm based on the correlation between each face recognition algorithm and other face recognition algorithms in the preset multiple face recognition algorithms, and obtain the second annotation result of each face image to be annotated. It uses multiple face recognition algorithms and evaluates the quality of the face images to be annotated according to the correlation between the multiple face recognition algorithms, which can reduce the dependence on a single face recognition algorithm, obtain face annotation data with generalization ability, and solve the problem that the existing face annotation data generation method does not have generalization ability because it is overly dependent on the current face recognition algorithm.
[0042] Details of one or more embodiments of the present application are set forth in the following drawings and description to make other features, objects, and advantages of the present application more readily apparent. BRIEF DESCRIPTION OF THE DRAWINGS
[0043] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:
[0044] Figure 1 A hardware structure block diagram of a terminal for a method for generating face annotation data provided in an embodiment of the present application;
[0045] Figure 2 A flowchart of a method for generating face annotation data provided in an embodiment of the present application;
[0046] Figure 3 A flowchart of a method for generating face annotation data provided in a preferred embodiment of the present application;
[0047] Figure 4 This is a structural block diagram of a device for generating face annotation data provided in one embodiment of the present application. DETAILED DESCRIPTION
[0048] In order to more clearly understand the purpose, technical solutions and advantages of the present application, the present application is described and illustrated below in conjunction with the accompanying drawings and embodiments.
[0049] Unless otherwise defined, the technical terms or scientific terms involved in this application shall have the general meaning understood by people with general skills in the technical field to which this application belongs. The words "one", "a", "the", "these" and the like in this application do not indicate a quantitative limitation, and they may be singular or plural. The terms "include", "comprise", "have" and any variants thereof involved in this application are intended to cover non-exclusive inclusions; for example, a process, method and system, product or device comprising a series of steps or modules (units) is not limited to the listed steps or modules (units), but may include unlisted steps or modules (units), or may include other steps or modules (units) inherent to these processes, methods, products or devices. The words "connect", "connected", "coupled" and the like involved in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The "multiple" involved in this application refers to two or more. "And / or" describes the association relationship of associated objects, indicating that there may be three relationships, for example, "A and / or B" may mean: A exists alone, A and B exist at the same time, and B exists alone. Generally, the character " / " indicates that the objects associated with each other are in an "or" relationship. The terms "first", "second", "third", etc. in this application are only used to distinguish similar objects and do not represent a specific ordering of the objects.
[0050] The method embodiment provided in this embodiment can be executed in a terminal, a computer or a similar computing device. For example, running on a terminal, Figure 1 FIG. 1 is a hardware structure block diagram of a terminal of the method for generating face annotation data of this embodiment. Figure 1 As shown, the terminal may include one or more ( Figure 1 Only one is shown in the figure) a processor 102 and a memory 104 for storing data, wherein the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA. The above terminal may also include a transmission device 106 and an input and output device 108 for communication functions. It can be understood by those skilled in the art that Figure 1The structure shown is for illustration only and does not limit the structure of the above terminal. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations shown.
[0051] The memory 104 can be used to store computer programs, for example, software programs and modules of application software, such as the computer program corresponding to the method for generating face annotation data in this embodiment. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, that is, to implement the above method. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories may be connected to the terminal via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.
[0052] The transmission device 106 is used to receive or send data via a network. The above network includes a wireless network provided by the communication provider of the terminal. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, referred to as NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency (Radio Frequency, referred to as RF) module, which is used to communicate with the Internet wirelessly.
[0053] In this embodiment, a method for generating face annotation data is provided. Figure 2 is a flow chart of the method for generating face annotation data of this embodiment. Figure 2 As shown, the process includes the following steps:
[0054] Step S210, obtaining a face image to be annotated and a standard image corresponding to the face image to be annotated.
[0055] In this step, the facial images to be annotated may be multiple image frames of multiple faces used to generate valid facial annotation data. The facial images to be annotated may be multiple image data in one or more formats such as bmp, jpg, png, tif, gif, etc. The facial images to be annotated may be image sequences obtained based on video sequences or image sets of different people. Among them, the standard image may be an ID photo of the person corresponding to the facial image to be annotated, or a standard photo of the person corresponding to the facial image to be annotated. It should be noted that the standard photo may be a picture determined by the user to be used as a standard. In this step, by acquiring the facial images to be annotated and the standard images corresponding to the facial images to be annotated, it is convenient to perform quality assessment on the facial images to be annotated according to the standard images corresponding to the facial images to be annotated, using a plurality of preset face recognition algorithms, so as to obtain the first annotation results of each facial image to be annotated for each face recognition algorithm.
[0056] Step S220, using a plurality of preset face recognition algorithms, similarity calculation is performed on each face image to be labeled and the face features in the corresponding standard image, to obtain a first labeling result of each face image to be labeled for each face recognition algorithm.
[0057] Among them, the above-mentioned multiple face recognition algorithms can be face recognition algorithms for extracting facial features from images and performing quality assessment on face images based on the facial features. Specifically, it can be a face recognition algorithm that extracts facial features from the face image to be labeled and the standard image respectively, and performs similarity assessment on the face image to be labeled and the standard image based on the facial features of the face image to be labeled and the facial features of the standard image. The above-mentioned preset multiple face recognition algorithms can be a method of performing face recognition using multiple face recognition models, for example, a method of performing face recognition using several face recognition models such as VGG-Face, GoogleFaceNet, OpenFace, Facebook DeepFace, DeepID, ArcFace, Dlib and SFace. The above-mentioned first annotation result can be the quality assessment result of each face image to be labeled for each face recognition algorithm, which can specifically be a score value.
[0058] It should be noted that, in the process of using multiple preset face recognition algorithms to calculate the similarity of each face image to be labeled and the face features in the corresponding standard image, and obtaining the first labeling result of each face image to be labeled for each face recognition algorithm, the following two steps can be included: the first step is to use the above-mentioned multiple face recognition algorithms to perform feature extraction on each face image to be labeled and the standard image corresponding to each face image to be labeled, and obtain the face features of each face image to be labeled and the face features of the standard image corresponding to each face image to be labeled; the second step is to calculate the similarity of the face features of each face image to be labeled and the face features of the standard image corresponding to each face image to be labeled, and obtain the quality evaluation result of the face image to be labeled for each face recognition algorithm, that is, the first labeling result.
[0059] Among them, in the above-mentioned first step, during the process of extracting facial features, it is necessary to weaken the influence of the alignment algorithm as much as possible. For example, when using the above-mentioned face recognition model to extract facial features, only the necessary binocular alignment is used or the alignment algorithm that depends on feature points is not used, but the facial features in the image are directly extracted. The method of directly extracting facial features from the image can minimize the influence of the alignment algorithm on the quality assessment results of the facial images to be annotated, that is, it can weaken the influence of over-correction on the quality assessment results of the facial images to be annotated. This makes the first annotation result obtained more generalizable.
[0060] Furthermore, in the above-mentioned second step, this embodiment evaluates the quality of the facial images to be labeled by calculating the similarity of the feature points of the facial images to be labeled and the standard images corresponding to each facial image to be labeled. The higher the similarity, the higher the quality of the facial images to be labeled, and the lower the similarity, the lower the quality of the facial images to be labeled.
[0061] It should be noted that different face recognition algorithms use different similarity measurement methods. Generally, the Euclidean distance or cosine distance between facial features can be used to measure the similarity between the face image to be labeled and the standard image. If the face recognition algorithm uses the cosine distance between facial features to measure the similarity between the face image to be labeled and the standard image, the similarity value between the face image to be labeled and the standard image can be directly used as the first labeling result of the face image to be labeled for the current face recognition algorithm. If the face recognition algorithm uses the Euclidean distance between facial features to measure the similarity between the face image to be labeled and the standard image, the Euclidean distance between facial features can be corrected, and the Euclidean distance can be mapped to the interval range of [0, 1] to generate a similarity representation between the face image to be labeled and the standard image, and the similarity representation between the face image to be labeled and the standard image can be used as the first labeling result of the face image to be labeled for the current face recognition algorithm. The Euclidean distance between the facial features may be corrected by using a sigmoid function to correct the Euclidean distance between the facial features, and mapping the Euclidean distance to an interval range of [0, 1].
[0062] Step S230, based on the correlation between each face recognition algorithm and other face recognition algorithms in the preset multiple face recognition algorithms, weighted calculation is performed on the first labeling results of each face recognition algorithm for each face image to be labeled, so as to obtain a second labeling result for each face image to be labeled.
[0063] The correlation between each face recognition algorithm and other face recognition algorithms in the preset multiple face recognition algorithms can be obtained by using one or more correlation evaluation indicators such as SROCC (Spearman Rank Order Correlation Coefficient), KROCC (Kendall Rank Order Correlation Coefficient), etc. to obtain the correlation between each face recognition algorithm and other face recognition algorithms in the preset multiple face recognition algorithms. The weighted calculation of the first annotation results of each face recognition algorithm for each face image to be annotated based on the correlation between each face recognition algorithm and other face recognition algorithms in the preset multiple face recognition algorithms can be firstly based on the correlation between each face recognition algorithm and other face recognition algorithms in the preset multiple face recognition algorithms, the weighted coefficients of each face image to be annotated under each face recognition algorithm are obtained, and then, the weighted coefficients of each face image to be annotated under each face recognition algorithm are used to perform weighted calculation on the first annotation results of each face recognition algorithm for each face image to be annotated. This step measures the contribution of each face recognition algorithm to the quality evaluation of the face image to be labeled through the correlation between each face recognition algorithm and other face recognition algorithms in the preset multiple face recognition algorithms. Based on this, a weighted calculation is performed on the first labeling results of each face image to be labeled for each face recognition algorithm to obtain the contribution of different face recognition algorithms to the quality evaluation of the labeled face image, so that the second labeling result of the quality of the face to be labeled is more accurate and has better generalization ability.
[0064] In the above steps S210 to S230, by using a plurality of preset face recognition algorithms, similarity calculation is performed on each face image to be labeled and the face features in the standard image corresponding to the face image to be labeled, and the first labeling result of each face image to be labeled for each face recognition algorithm is obtained. Then, based on the correlation between each face recognition algorithm and other face recognition algorithms in the preset plurality of face recognition algorithms, weighted calculation is performed on the first labeling result of each face image to be labeled for each face recognition algorithm, and the second labeling result of each face image to be labeled is obtained. By using a plurality of face recognition algorithms and evaluating the quality of the face images to be labeled according to the correlation between the plurality of face recognition algorithms, the dependence on a single face recognition algorithm can be reduced, and face labeling data with generalization ability can be obtained, which solves the problem that the existing face labeling data generation method does not have generalization ability because it is overly dependent on the current face recognition algorithm.
[0065] In one embodiment, the above method for generating face annotation data further includes the following steps:
[0066] Step S240, filtering abnormal data in the second annotation results of each face image to be annotated to obtain valid face annotation data.
[0067] In generating the second annotation result, there may be a large gap between the second annotation results of very similar images, which may cause confusion in the data segment, making the face quality assessment model trained using the second annotation result inaccurate. There may also be similar images that appear multiple times, and too many similar images will bring certain uncertainties to the subsequent training and testing of the face quality assessment model. This requires filtering the abnormal data in the second annotation results of the face image to be annotated to obtain effective face annotation data. The above-mentioned abnormal data can be the large data gap between the second annotation results of very similar images, can be multiple redundant data corresponding to too many similar images, and can be data corresponding to the situation where the face image to be annotated after being sorted according to the second annotation result has a neighborhood anomaly. This embodiment filters the abnormal data in the second annotation results of each face image to be annotated, and obtains effective face annotation data, which can use the effective face annotation data to train the face quality assessment model, so that the training effect of the complete face quality assessment model is better.
[0068] In addition, in one embodiment, the above step S220 uses a plurality of preset face recognition algorithms to perform similarity calculations on the face features in each face image to be labeled and the corresponding standard image to obtain a first labeling result of each face image to be labeled for each face recognition algorithm, including the following steps:
[0069] Step S222, using a plurality of preset face recognition algorithms, similarity calculations are performed on the face features in each face image to be labeled and the corresponding standard image, to obtain similarity results of each face image to be labeled under each face recognition algorithm.
[0070] Step S224, the similarity results of each face image to be labeled under each face recognition algorithm are expressed in the same measurement method to obtain the first labeling results of each face image to be labeled for each face recognition algorithm.
[0071] It should be noted that, because different face recognition algorithms use different similarity measurement methods, it is necessary to use the same measurement method to represent the similarity results of each face image to be labeled under each face recognition algorithm, so that the first labeling result is more accurate. The above-mentioned same measurement method can represent the similarity result in percentage, and can also map the similarity result to the range of [0, 1] for representation. This embodiment does not specifically limit the measurement method, as long as it can ensure that the first labeling result of each face image to be labeled for each face recognition algorithm uses the same measurement method.
[0072] In the above steps S222 to S224, by utilizing the preset multiple face recognition algorithms, similarity calculations are performed on the face features in each face image to be labeled and the corresponding standard image, so as to obtain the similarity results of each face image to be labeled under each face recognition algorithm, and the similarity results are expressed in the same measurement method, so as to obtain the first labeling results of each face image to be labeled for each face recognition algorithm, thereby ensuring the accuracy of the first labeling results.
[0073] Further, in one embodiment, the above step S230, based on the correlation between each face recognition algorithm and other face recognition algorithms in the preset multiple face recognition algorithms, performs weighted calculation on each face image to be labeled for the first labeling result of each face recognition algorithm to obtain the second labeling result of each face image to be labeled, includes the following steps:
[0074] Step S232, based on the correlation between each face recognition algorithm and other face recognition algorithms in the preset multiple face recognition algorithms, obtain the weighting coefficient of each face image to be labeled under each face recognition algorithm.
[0075] The correlation between each face recognition algorithm and other face recognition algorithms in the preset multiple face recognition algorithms can be the correlation between two face recognition algorithms obtained by comparing each face recognition algorithm with the face recognition algorithms other than the current face recognition algorithm in the preset multiple face recognition algorithms. Therefore, there are multiple correlation results corresponding to each face recognition algorithm. For example, if there are N face recognition algorithms in total, each face recognition algorithm corresponds to N-1 correlation results. Furthermore, the weighting coefficients of each face image to be labeled under each face recognition algorithm can be obtained by calculating the mean of the correlation between each face recognition algorithm and other face recognition algorithms in the preset multiple face recognition algorithms.
[0076] Step S234, based on the weighted coefficients of each face image to be labeled under each face recognition algorithm, weighted calculation is performed on each face image to be labeled for the first labeling result of each face recognition algorithm to obtain a second labeling result of each face image to be labeled.
[0077] In this step, the weighted calculation is performed on the first annotation results of each face recognition algorithm for each face image to be annotated based on the weighted coefficients of each face image to be annotated under each face recognition algorithm to obtain the second annotation results of each face image to be annotated. The following operations can be performed on each face image to be annotated: multiplying the first annotation results of each face recognition algorithm by the weighted coefficient of the current face image to be annotated under the current face recognition algorithm to obtain the weighted annotation results corresponding to each face recognition algorithm; summing the weighted annotation results corresponding to each face recognition algorithm to obtain the second annotation result of the current face image to be annotated. The second annotation result of each face image to be annotated is obtained by the above method, and the contribution of different face recognition algorithms to the quality evaluation of the face image to be annotated can be combined, so that the second annotation result of the quality of the face to be annotated is more accurate.
[0078] In the above steps S232 to S234, based on the correlation between each face recognition algorithm and other face recognition algorithms among the preset multiple face recognition algorithms, weighted coefficients of each face image to be labeled under each face recognition algorithm are obtained, and then based on the weighted coefficients of each face image to be labeled under each face recognition algorithm, weighted calculation is performed on the first labeling results of each face image to be labeled for each face recognition algorithm, and the contribution of each face recognition algorithm to the quality evaluation of the labeled face image is measured through the correlation between each face recognition algorithm and other face recognition algorithms among the preset multiple face recognition algorithms. Because the contribution of different face recognition algorithms to the quality evaluation of the labeled face image is fully considered, the second labeling result of the quality of the face to be labeled is more accurate and has better generalization ability.
[0079] Specifically, in one embodiment, the above step S232, based on the correlation between each face recognition algorithm and other face recognition algorithms in the preset multiple face recognition algorithms, obtains the weighting coefficient of each face image to be labeled under each face recognition algorithm, including the following steps:
[0080] Step S2322, based on the correlation between each face recognition algorithm and other face recognition algorithms in the preset multiple face recognition algorithms, obtain the average value of the correlation between each face recognition algorithm and other face recognition algorithms in the preset multiple face recognition algorithms.
[0081] Step S2324, normalizing the mean of the correlation between the current face recognition algorithm and other face recognition algorithms in the preset multiple face recognition algorithms as the weighting coefficient of each face image to be labeled under the current face recognition algorithm.
[0082] The above-mentioned calculation of the normalized result of the mean of the correlation between the current face recognition algorithm and other face recognition algorithms in the preset multiple face recognition algorithms can be to first calculate the mean of the correlation between each face recognition algorithm and other face recognition algorithms in the preset multiple face recognition algorithms to obtain the correlation mean corresponding to each face recognition algorithm, and then calculate the sum of the correlation means corresponding to all face recognition algorithms, and then calculate the correlation mean corresponding to the current face recognition algorithm divided by the sum of the correlation means, and use the obtained quotient as the normalized result of the mean of the correlation between the current face recognition algorithm and other face recognition algorithms in the preset multiple face recognition algorithms. That is, the correlation mean corresponding to the current face recognition algorithm is divided by the sum of the correlation means corresponding to all face recognition algorithms, and the obtained quotient is used as the weighting coefficient of each face image to be labeled under the current face recognition algorithm.
[0083] In the above steps S2322 to S2324, based on the correlation between each face recognition algorithm and other face recognition algorithms in the preset multiple face recognition algorithms, the mean of the correlation between each face recognition algorithm and other face recognition algorithms in the preset multiple face recognition algorithms is obtained, and the normalized result of the mean of the correlation between the current face recognition algorithm and other face recognition algorithms in the preset multiple face recognition algorithms is obtained, and the normalized result of the mean of the correlation between the current face recognition algorithm and other face recognition algorithms in the preset multiple face recognition algorithms is used as the weighting coefficient of each face image to be labeled under the current face recognition algorithm. The importance of each current face recognition algorithm in the contribution to the image sequence ranking of the current face image to be labeled is reflected through the normalized result of the mean of the correlation between the current face recognition algorithm and other face recognition algorithms in the preset multiple face recognition algorithms, and the second labeling result of the quality of the face to be labeled is made more accurate by fully considering the contribution of different face recognition algorithms to the ranking of the image sequence.
[0084] In one embodiment, after obtaining the second annotation result of each face image to be annotated, the method further includes:
[0085] Step S250 , based on the preset absolute quality scores of the facial images to be labeled, and in accordance with the preset step length, segmentally map the second labeling results of the facial images to be labeled to obtain third labeling results of the facial images to be labeled.
[0086] When there are different people whose face images have the same second annotation results, but the image quality of the face images with the same second annotation results is not uniform, the second annotation results need to be further adjusted to avoid the above situation. The above-mentioned preset absolute quality scores of each face image to be annotated can be the manual annotation results of each face image to be annotated by the user. The above-mentioned preset step size can be set according to the specific situation. For example, if the range of the second annotation result is [0, 1], the preset step size can be set to 0.1. The above-mentioned second annotation results of each face image to be annotated are segmented and mapped according to the preset step size. Specifically, it can be: according to the preset step size, the range of the second annotation result is divided into multiple quantization orders; the absolute quality scores of each face image to be annotated are divided into each quantization order; all the absolute quality scores in each quantization order are averaged to obtain the average value corresponding to each quantization order; based on the average value corresponding to each quantization order, the second annotation results of each face image to be annotated are segmented and mapped.
[0087] Step S260, filtering abnormal data in the third annotation results of each face image to be annotated to obtain valid face annotation data.
[0088] In the above steps S250 to S260, the second annotation results of each facial image to be annotated are segmented and mapped according to a preset step size to obtain the third annotation results of each facial image to be annotated, and then the abnormal data in the third annotation results of each facial image to be annotated are filtered to obtain effective facial annotation data, which solves the problem that the second annotation results of facial images of different people are the same, but the image quality of facial images with the same second annotation results are not uniform, so that the third annotation results are more accurate, and then more accurate effective facial annotation data are obtained, so that the facial quality assessment model is trained through the effective facial annotation data, so that the training effect of the complete facial quality assessment model is better.
[0089] In one embodiment, the above step S240, filtering abnormal data in the second annotation results of each face image to be annotated to obtain valid face annotation data, includes the following steps:
[0090] Step S242: clustering the facial images to be labeled according to facial features to obtain clustering results.
[0091] In general, when there are very similar facial images to be labeled but the second labeling results are quite different, the very similar facial images to be labeled can be clustered, and the data in the same cluster after clustering can be screened to remove abnormal data. The above-mentioned clustering of facial images to be labeled based on facial features can be performed by clustering the facial images to be labeled using one or more clustering methods such as K-means clustering, hierarchical clustering, DBSCAN (Density-Based Spatial Clustering of Applications with Noise, density-based clustering method with noise). Specifically, the facial images to be labeled can be clustered using one facial feature, or by using multiple facial features. The above-mentioned clustering results are multiple clusters obtained by grouping multiple facial images to be labeled.
[0092] It should be noted that the purpose of this clustering is to remove highly similar face images to be labeled, not to roughly classify the face images to be identified according to faces. Therefore, the number of clusters can be set to a relatively large value. For example, the number of clusters can be set between half of the total number of sequences and the total number of sequences. The above total number of sequences can be the number of image sequences that use the face images of the same person as a sequence. This step can cluster the face images to be labeled according to facial features, group very similar images into the same cluster, and then select a part of the samples from each cluster as a data set of valid face labeling data. In this way, it can be ensured that the samples in the generated data set of valid face labeling data have certain differences, avoid the situation where too many highly similar face images to be labeled lead to confusion of data segments, and improve the accuracy and reliability of face labeling data.
[0093] Step S244, based on the number of facial images to be labeled in each cluster of the clustering result and the second labeling result of each facial image to be labeled, the facial images to be labeled in each cluster are simplified to obtain simplified facial images to be labeled.
[0094] The above-mentioned method of simplifying the face images to be labeled in each cluster based on the number of face images to be labeled in each cluster based on the clustering results and the second labeling results of each face image to be labeled can be: when the number of face images to be labeled in each cluster based on the clustering results is 1, the face images to be labeled in the current cluster and the second labeling results of the face images to be labeled are retained; when the number of face images to be labeled in a cluster is greater than or equal to 2, the face images to be labeled in each cluster are simplified based on the difference between the second labeling results of each face image to be labeled in the cluster.
[0095] Step S246, sorting the streamlined face images to be labeled according to the second labeling results, and performing neighborhood anomaly filtering on the second labeling results of the sorted face images to be labeled to obtain valid face labeling data.
[0096] Among them, the streamlined facial images to be annotated are sorted according to the second annotation result, which can be sorting the facial images to be annotated in order of score values from large to small, or in order of score values from small to large. The above-mentioned neighborhood anomaly filtering can include feature anomaly filtering and attribute anomaly filtering. The above-mentioned feature anomaly filtering can be obtained by calculating the similarity of adjacent images and interval images, and the similarity results of adjacent images and interval images are obtained. Based on the similarity results of adjacent images and interval images, abnormal data is determined, and the abnormal data and images corresponding to the abnormal data are filtered. The above-mentioned attribute anomaly filtering can be based on the attribute scores of adjacent images, determine the difference in attribute scores of adjacent images, determine the abnormal data based on the difference in attribute scores of adjacent images, and filter the abnormal data and images corresponding to the abnormal data.
[0097] In the above steps S242 to S246, the facial images to be labeled are clustered, and based on the number of facial images to be labeled in each cluster of the clustering results, the facial images to be labeled in each cluster are simplified, and the simplified facial images to be labeled are subjected to neighborhood anomaly filtering to obtain valid facial annotation data. By filtering the abnormal data in the second annotation results of each facial image to be labeled, the valid facial annotation data obtained can ensure that the facial quality assessment model is trained using the valid facial annotation data, so that the training effect of the complete facial quality assessment model is better.
[0098] Furthermore, the above step S244, based on the number of facial images to be labeled in each cluster of the clustering result and the second labeling result of each facial image to be labeled, simplifies the facial images to be labeled in each cluster to obtain simplified facial images to be labeled, including the following steps:
[0099] Step S2442: when the number of the facial images to be labeled in the cluster is greater than or equal to 2, the difference between the second labeling results of the facial images to be labeled in the cluster is calculated.
[0100] Step S2444: based on the difference between the second annotation results of each facial image to be annotated in the cluster, simplify the facial images to be annotated in each cluster to obtain simplified facial images to be annotated.
[0101] In this step, the facial images to be labeled in each cluster are simplified based on the difference between the second annotation results of each facial image to be labeled in the cluster. The maximum difference between the second annotation results of the facial images to be labeled in the current cluster is determined based on the difference between the second annotation results of each facial image to be labeled in the cluster; if the maximum difference between the second annotation results of the facial images to be labeled in the current cluster is less than or equal to the first difference threshold, the middle position of the second annotation results of the facial images to be labeled in the current cluster after sorting is determined (if there are two images in the middle position, either image can be selected). A face image to be annotated (such as an image in the middle position) is used as the middle image, and other face images to be annotated in the current cluster except the middle image are deleted; if the maximum difference between the second annotation results of the face images to be annotated in the current cluster is greater than or equal to the second difference threshold, all face images to be annotated in the current cluster are deleted; if the maximum difference between the second annotation results of the face images to be annotated in the current cluster is between the first difference threshold and the second difference threshold, a representative image and the second annotation result corresponding to the representative image are selected by using a preset selection strategy, and images except the representative image in the current cluster are deleted. The above-mentioned first difference threshold and second difference threshold can be specifically set according to actual needs. Moreover, the first difference threshold is less than the second difference threshold. If the maximum difference between the second annotation results of the facial images to be annotated in the current cluster is less than or equal to the first difference threshold, it is determined that the annotation scores of all images in the current cluster are very close, that is, there are multiple image frames to be identified with similar quality. At this time, in order to avoid excessive data redundancy caused by too many similar images and similar annotation results, only one image and the second annotation result corresponding to the above image need to be retained to achieve the simplification of the facial images to be annotated in each cluster. If the maximum difference between the second annotation results of the facial images to be annotated in the current cluster is greater than the second difference threshold, it is determined that the annotation scores of all images in the current cluster differ too much, and the data in the cluster is considered unreliable, and all data in the current cluster can be discarded.
[0102] The above-mentioned method of selecting the representative image and the second annotation result corresponding to the representative image by using the preset selection strategy may be to calculate the average value of the similarity between each face image to be annotated and other face images to be annotated in the current cluster, determine the face image to be annotated with the highest average value of similarity, and delete the face images to be annotated in the current cluster except the face image to be annotated with the highest average value of similarity. It may be to remove the face images to be annotated corresponding to the maximum score and the minimum score in the second annotation results in the current cluster, calculate the average value of the remaining second annotation results, select the second annotation result closest to the average value and the face image to be annotated corresponding to the second annotation result closest to the average value as the second annotation result and the representative image corresponding to the representative image of the current cluster. If there are two images in the current cluster, calculate the average value of the second annotation results of the two images, select the second annotation result closest to the average value and the face image to be annotated corresponding to the second annotation result closest to the average value as the second annotation result and the representative image corresponding to the representative image of the current cluster.
[0103] In the above steps S2442 to S2444, when the number of facial images to be labeled in a cluster is greater than or equal to 2, the difference between the second annotation results of each facial image to be labeled in the cluster is calculated, and the facial images to be labeled in each cluster are simplified according to the difference between the second annotation results of each facial image to be labeled in the cluster, so that unreliable clusters and redundant data in each cluster are eliminated, so as to avoid the situation where some data segments are confused due to a large difference in the second annotation results between very similar images, so that the simplified facial images to be labeled and the second annotation results of the facial images to be labeled are more accurate.
[0104] In addition, the above step S246, sorting the streamlined face images to be annotated according to the second annotation results, and performing neighborhood anomaly filtering on the second annotation results of the sorted face images to be annotated to obtain valid face annotation data, includes the following steps:
[0105] Step S2462, sorting the streamlined face images to be labeled according to the second labeling result, and calculating the similarity of adjacent images and interval images for the facial features in the sorted face images to be labeled, to obtain the similarity results of adjacent images and the similarity results of interval images.
[0106] Step S2464, based on the similarity results of adjacent images and the similarity results of interval images, feature anomaly filtering is performed on the streamlined face image to obtain a face image to be identified after feature anomaly filtering.
[0107] The following is a specific example of the process of filtering feature anomalies for streamlined face images based on the similarity results of adjacent images and the similarity results of interval images. For example, there are three adjacent face images, namely image A, image B, and image C. The similarity between any two images is calculated. When the similarity between image A and image B is less than the similarity between image A and image C, or the similarity between image C and image B is less than the similarity between image A and image C, it is considered that the existence of image B is unreasonable, and the data corresponding to image B is abnormal data. Image B and the second annotation result corresponding to image B need to be removed.
[0108] Step S2466, based on the difference in attribute scores between adjacent facial images to be identified after feature anomaly filtering, and a preset attribute score difference threshold, attribute anomaly filtering is performed on the facial images to be identified after feature anomaly filtering to obtain valid facial annotation data.
[0109] The above attributes may be attributes such as clarity, angle, and occlusion of the face image. Before performing attribute anomaly filtering, it is necessary to determine the attribute scores of the face images to be identified after filtering out the feature anomalies. Specifically, the attribute scores of the face images to be identified after filtering out the feature anomalies may be determined by using a plurality of preset face recognition algorithms. The above preset attribute score difference threshold may be specifically set according to the specific situation.
[0110] If the difference in attribute features between adjacent facial images to be identified after feature anomaly filtering is significantly greater than a set threshold, it indicates that the attribute distribution of these image data is inconsistent with the overall data set, and there may be low quality or other problems, and the abnormal data can be removed. Based on this, the above-mentioned attribute anomaly filtering is performed on the facial images to be identified after feature anomaly filtering based on the difference in attribute scores between adjacent facial images to be identified after feature anomaly filtering, and the preset attribute score difference threshold. The difference in attribute scores between adjacent facial images to be identified after feature anomaly filtering can be greater than the preset attribute score difference threshold, and it is determined that the quality assessment results of the adjacent facial images are unreasonable, and the second annotation results corresponding to the adjacent facial images are abnormal data.
[0111] In the above steps S2462 to S2466, the simplified face images to be annotated are sorted according to the second annotation results, the similarity results of adjacent images and the similarity results of interval images are calculated, and then, based on the similarity results of adjacent images and the similarity results of interval images, the simplified face images are subjected to feature anomaly filtering and attribute anomaly filtering to obtain effective face annotation data. By filtering the abnormal data in the second annotation results of each face image to be annotated, the effective face annotation data obtained can be used to train the face quality assessment model by using the effective face annotation data, so that the training effect of the complete face quality assessment model is better.
[0112] Preferably, the first face image of each image sequence can also be verified. Specifically, the first face image of each image sequence can be verified according to the attributes of the image. When the attribute score of the first face image is less than the preset minimum attribute threshold, it is determined that the first face image of the current image sequence does not meet the attribute requirements, and the first face image of the current image sequence is discarded. The above-mentioned minimum attribute threshold can be set according to the specific situation.
[0113] The present embodiment is described and illustrated below through preferred embodiments.
[0114] Figure 3 FIG. 1 is a flow chart of a method for generating face annotation data provided by a preferred embodiment of the present application. Figure 3 As shown, the method for generating face annotation data includes the following steps:
[0115] Step S310, obtaining a face image to be annotated and a standard image corresponding to the face image to be annotated;
[0116] Step S320, using a plurality of preset face recognition algorithms, similarity calculation is performed on each face image to be labeled and the face features in the corresponding standard image, and the similarity result of each face image to be labeled under each face recognition algorithm is obtained;
[0117] Step S330, expressing the similarity results of each face image to be labeled under each face recognition algorithm using the same measurement method, and obtaining the first labeling results of each face image to be labeled for each face recognition algorithm;
[0118] Step S340, based on the correlation between each face recognition algorithm and other face recognition algorithms in the preset multiple face recognition algorithms, obtain the weighting coefficient of each face image to be labeled under each face recognition algorithm;
[0119] Step S350, based on the weighted coefficients of each face image to be labeled under each face recognition algorithm, weighted calculation is performed on each face image to be labeled for the first labeling result of each face recognition algorithm to obtain a second labeling result of each face image to be labeled;
[0120] Step S360, clustering the facial images to be annotated according to facial features to obtain clustering results;
[0121] Step S370, based on the number of facial images to be labeled in each cluster of the clustering result and the second labeling result of each facial image to be labeled, simplifying the facial images to be labeled in each cluster to obtain simplified facial images to be labeled;
[0122] Step S380, sorting the streamlined face images to be labeled according to the second labeling results, and performing neighborhood anomaly filtering on the second labeling results of the sorted face images to be labeled to obtain valid face labeling data.
[0123] In the above steps S310 to S380, by using a plurality of preset face recognition algorithms, similarity calculation is performed on each face image to be labeled and the face features in the standard image corresponding to the face image to be labeled, and the first labeling result of each face image to be labeled for each face recognition algorithm is obtained. Then, based on the correlation between each face recognition algorithm and other face recognition algorithms in the preset plurality of face recognition algorithms, weighted calculation is performed on the first labeling result of each face image to be labeled for each face recognition algorithm, and the second labeling result of each face image to be labeled is obtained. By using a plurality of face recognition algorithms and evaluating the quality of the face images to be labeled according to the correlation between the plurality of face recognition algorithms, the dependence on a single face recognition algorithm can be reduced, and the face images to be labeled are abnormally filtered, and invalid face labeling data is simplified to obtain face labeling data with generalization ability, which solves the problem that the existing face labeling data generation method does not have generalization ability because it is overly dependent on the current face recognition algorithm.
[0124] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.
[0125] Based on the same inventive concept, a device for generating face annotation data is also provided in this embodiment, which is used to implement the above embodiments and preferred implementation modes, and will not be repeated hereafter. The terms "module", "unit", "subunit", etc. used below may be a combination of software and / or hardware that implements predetermined functions. Although the devices described in the following embodiments are preferably implemented in software, the implementation in hardware, or a combination of software and hardware, is also possible and conceivable.
[0126] In one embodiment, Figure 4 is a structural block diagram of a device for generating face annotation data provided by an embodiment of the present application, such as Figure 4 As shown, the device for generating face annotation data includes:
[0127] An acquisition module 42 is used to acquire a face image to be annotated and a standard image corresponding to the face image to be annotated;
[0128] The first calculation module 44 is used to calculate the similarity between each face image to be labeled and the face features in the corresponding standard image by using a plurality of preset face recognition algorithms, so as to obtain a first labeling result of each face image to be labeled for each face recognition algorithm;
[0129] And a second calculation module 46 is used to perform weighted calculation on the first annotation results of each face recognition algorithm for each face image to be annotated based on the correlation between each face recognition algorithm and other face recognition algorithms in the preset multiple face recognition algorithms, so as to obtain the second annotation results of each face image to be annotated.
[0130] The above-mentioned face annotation data generation device calculates the similarity between each face image to be annotated and the face features in the standard image corresponding to the face image to be annotated through a preset multiple face recognition algorithms, obtains the first annotation result of each face image to be annotated for each face recognition algorithm, and then performs weighted calculation on the first annotation result of each face image to be annotated for each face recognition algorithm based on the correlation between each face recognition algorithm and other face recognition algorithms in the preset multiple face recognition algorithms, and obtains the second annotation result of each face image to be annotated. By using multiple face recognition algorithms and evaluating the quality of the face images to be annotated according to the correlation between the multiple face recognition algorithms, it can reduce the dependence on a single face recognition algorithm, obtain face annotation data with generalization ability, and solve the problem that the existing face annotation data generation method does not have generalization ability because it is overly dependent on the current face recognition algorithm.
[0131] It should be noted that the above modules can be functional modules or program modules, and can be implemented by software or hardware. For modules implemented by hardware, the above modules can be located in the same processor; or the above modules can be located in different processors in any combination.
[0132] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, any one of the methods for generating face annotation data in the above embodiments is implemented.
[0133] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.
[0134] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0135] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.
Claims
1. A method for generating face annotation data, characterized in that: The method comprises: Acquire a face image to be annotated and a standard image corresponding to the face image to be annotated; Using a plurality of preset face recognition algorithms, similarity calculation is performed on each of the face images to be labeled and the face features in the corresponding standard image, so as to obtain a first labeling result of each of the face images to be labeled for each face recognition algorithm; Based on the correlation between each face recognition algorithm and other face recognition algorithms among the preset multiple face recognition algorithms, weighted calculation is performed on each of the face images to be labeled for the first labeling results of each face recognition algorithm to obtain a second labeling result for each of the face images to be labeled; Perform neighborhood anomaly filtering on the abnormal data in the second annotation results of each of the face images to be annotated to obtain valid face annotation data; the neighborhood anomaly filtering includes feature anomaly filtering and attribute anomaly filtering; the feature anomaly filtering is to determine and filter the abnormal data through the similarity results of adjacent images and the similarity results of interval images; the attribute anomaly filtering is to determine and filter the abnormal data through the difference in attribute scores of the adjacent images.
2. The method for generating face annotation data according to claim 1, characterized in that: The method of using a plurality of preset face recognition algorithms to perform similarity calculation on each of the face images to be labeled and the face features in the corresponding standard image to obtain a first labeling result of each of the face images to be labeled for each face recognition algorithm includes: Using the preset multiple face recognition algorithms, similarity calculation is performed on each of the face images to be labeled and the face features in the corresponding standard image to obtain similarity results of each of the face images to be labeled under each of the face recognition algorithms; The similarity results of each of the facial images to be labeled under each of the facial recognition algorithms are expressed in the same measurement method to obtain the first labeling results of each of the facial images to be labeled for each of the facial recognition algorithms.
3. The method for generating face annotation data according to claim 1, characterized in that: The step of performing weighted calculation on each of the face images to be labeled for the first labeling results of each of the face recognition algorithms based on the correlation between each of the face recognition algorithms and other face recognition algorithms among the preset multiple face recognition algorithms to obtain a second labeling result for each of the face images to be labeled includes: Based on the correlation between each face recognition algorithm and other face recognition algorithms in the preset multiple face recognition algorithms, a weighting coefficient of each face image to be labeled under each face recognition algorithm is obtained; Based on the weighted coefficients of each of the facial images to be labeled under each of the facial recognition algorithms, weighted calculations are performed on each of the facial images to be labeled for the first labeling results of each of the facial recognition algorithms to obtain second labeling results for each of the facial images to be labeled.
4. The method for generating face annotation data according to claim 3, characterized in that: The step of obtaining weighted coefficients of each face image to be labeled under each face recognition algorithm based on the correlation between each face recognition algorithm and other face recognition algorithms in the preset multiple face recognition algorithms includes: Based on the correlation between each face recognition algorithm and other face recognition algorithms in the preset multiple face recognition algorithms, obtaining the average value of the correlation between each face recognition algorithm and other face recognition algorithms in the preset multiple face recognition algorithms; The normalized result of the mean value of the correlation between the current face recognition algorithm and other face recognition algorithms in the preset multiple face recognition algorithms is used as the weighting coefficient of each face image to be labeled under the current face recognition algorithm.
5. The method for generating face annotation data according to claim 1, characterized in that: After obtaining the second annotation result of each of the face images to be annotated, the method further includes: Based on the preset absolute quality scores of the facial images to be labeled, and in accordance with the preset step length, segmentally mapping the second labeling results of the facial images to be labeled, to obtain third labeling results of the facial images to be labeled; Abnormal data in the third annotation result of each of the face images to be annotated is filtered to obtain the valid face annotation data.
6. The method for generating face annotation data according to claim 1, characterized in that: The performing neighborhood anomaly filtering on the abnormal data in the second annotation results of each of the face images to be annotated to obtain valid face annotation data includes: Clustering the face images to be labeled according to face features to obtain clustering results; Based on the number of facial images to be labeled in each cluster of the clustering result and the second labeling result of each facial image to be labeled, simplifying the facial images to be labeled in each cluster to obtain simplified facial images to be labeled; The streamlined face images to be annotated are sorted according to the second annotation results, and the neighborhood anomaly filtering is performed on the sorted second annotation results of the face images to be annotated to obtain the valid face annotation data.
7. The method for generating face annotation data according to claim 6, characterized in that: The method of simplifying the facial images to be labeled in each cluster based on the number of facial images to be labeled in each cluster of the clustering result and the second labeling result of each facial image to be labeled to obtain simplified facial images to be labeled includes: When the number of the facial images to be labeled in the cluster is greater than or equal to 2, calculating the difference between the second labeling results of each of the facial images to be labeled in the cluster; Based on the difference between the second annotation results of each of the facial images to be annotated in the cluster, the facial images to be annotated in each of the clusters are simplified to obtain the simplified facial images to be annotated.
8. The method for generating face annotation data according to claim 6, characterized in that: The step of sorting the streamlined face images to be labeled according to the second labeling results, and performing the neighborhood anomaly filtering on the sorted second labeling results of the face images to be labeled to obtain the valid face labeling data includes: Sorting the streamlined face images to be annotated according to the second annotation result, and calculating the similarity of adjacent images and interval images for the facial features in the sorted face images to be annotated, to obtain the similarity results of the adjacent images and the similarity results of the interval images; Based on the similarity results of the adjacent images and the similarity results of the interval images, the simplified face image is subjected to the feature abnormality filtering to obtain the face image to be identified after the feature abnormality filtering; Based on the difference in attribute scores between adjacent facial images to be identified after feature anomaly filtering and a preset attribute score difference threshold, the facial images to be identified after feature anomaly filtering are subjected to attribute anomaly filtering to obtain the valid facial annotation data.
9. A device for generating face annotation data, characterized in that: The device comprises: An acquisition module, used to acquire a face image to be annotated and a standard image corresponding to the face image to be annotated; A first calculation module is used to use a plurality of preset face recognition algorithms to perform similarity calculations on the face features in each of the face images to be labeled and the corresponding standard images, so as to obtain a first labeling result of each of the face images to be labeled for each face recognition algorithm; And a second calculation module is used to perform weighted calculation on the first annotation results of each face recognition algorithm for each face image to be annotated based on the correlation between each face recognition algorithm and other face recognition algorithms in the preset multiple face recognition algorithms, so as to obtain a second annotation result of each face image to be annotated; perform neighborhood anomaly filtering on the abnormal data in the second annotation result of each face image to be annotated, so as to obtain valid face annotation data; the neighborhood anomaly filtering includes feature anomaly filtering and attribute anomaly filtering; the feature anomaly filtering is to determine and filter the abnormal data through the similarity results of adjacent images and the similarity results of interval images; the attribute anomaly filtering is to determine and filter the abnormal data through the difference of the attribute scores of the adjacent images.
10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the method for generating face annotation data according to any one of claims 1 to 8 are implemented.
